Evaluation models of root biomass in forest stands

David Galvagni, Enrico Gregori, Giovanni Zorn

Abstract


A comparison between two different approaches for forest root biomass estimation from some commonly surveyed dendrometrical variables is performed. The analytical set included data of 264 stands collected from the literature, but only 180 records covered the whole range of variables: age, basimetric area, density, above ground and underground biomass.
Results obtained by regression analysis (forward stepwise multiple regression) and by neural networks were compared after separation of dataset in two clusters (softwoods and hardwoods). Two subsets of observations were further established in a completely random way in order to build the models and to test their predictive performances; for this last issue a sample representing 20 and 27% of the observations was selected for softwoods and hardwoods respectively. The capabilities of the models to fit the relationships between biomasses and dendrometric variables as well as to predict them from stand measurements resulted very similar following both approaches; nevertheless, neural networks showed a better level of predictive accuracy. Anyway, the uncertainty remains considerable, with standard error of estimates never lower than 28% and 25% of the mean value for root biomass and above ground biomass respectively. The biomass ratio between root- and aboveground apparatus averaged 0.24, without any difference between softwoods and hardwoods.

Keywords


forest stand biomass; roots; estimation functions; neural networks; multiple regression analysis

Full Text: PDF (Italiano)

Refbacks

  • There are currently no refbacks.